GPU Performance Engineer (CUDA / AI Workloads)

About the Role

We’re looking for a GPU Performance Engineer to optimize and scale high-performance AI workloads at the kernel level.

This role focuses on deep performance engineering — working directly with CUDA kernels, GPU architecture, and profiling tools to unlock efficiency across training and inference systems. You’ll work on compute-intensive workloads where every microsecond and memory access pattern matters.

If you enjoy low-level optimization, understanding how hardware really works, and pushing systems to their limits — this role is for you.

What You’ll Do

• Design, optimize, and debug CUDA kernels for high-performance AI workloads
• Improve performance of compute-intensive operations (e.g., GEMM, attention, MoE, graph workloads)
• Apply low-level optimization techniques across memory hierarchy, warp execution, tensor cores, and compute efficiency
• Profile and eliminate GPU bottlenecks using tools like Nsight Systems and Nsight Compute
• Integrate optimized kernels into modern AI training and inference stacks
• Build internal tooling for benchmarking, profiling, and performance analysis
• Collaborate with engineering teams on performance-critical model infrastructure
• Contribute reusable performance improvements and best practices across the stack

What We’re Looking For

• Strong hands-on experience with CUDA kernel development and optimization
• Deep understanding of GPU architecture (memory hierarchy, warp scheduling, synchronization)
• Strong C/C++ skills for high-performance systems programming
• Proven experience identifying and resolving GPU performance bottlenecks
• Experience optimizing compute-heavy workloads in real-world systems
• Familiarity with GPU profiling tools (Nsight Systems, Nsight Compute, or similar)
• Experience integrating custom kernels into ML frameworks
• Strong debugging skills and performance-oriented mindset

Nice to Have

• Experience with PTX and architecture-specific optimizations
• Familiarity with modern GPU architectures (e.g., Hopper, Blackwell)
• Experience optimizing Transformer workloads (e.g., attention, FlashAttention)
• Familiarity with Triton, CUTLASS, Thrust, or CUB
• Experience with multi-GPU systems (e.g., NVLink, NCCL)
• Open-source contributions or relevant research in GPU / HPC

Why This Role

• Work on cutting-edge AI systems with real performance impact
• Solve complex low-level engineering problems at the hardware boundary
• High ownership over performance-critical components
• Opportunity to push modern GPU systems to their limits